// RUN — FITS

Can the Apple M3 Max 36GB run LLaVA-CoT?

Yes — here's how fast and up to what context, with real measured numbers.

LLaVA-CoT on the Apple M3 Max 36GB

QuantFits?Max contextSpeedBeyond context
FP16 yes 54k 9.3-11.2 t/s ~0.3-0.4 (slow)
Q8 yes 121k 18.5-22.4 t/s ~0.6-0.7 (slow)
Q4 yes 151k 33.1-39.9 t/s ~1.1-1.3 (slow)

Estimate (memory-bound). Beyond "max context" the KV cache spills to SSD swap and speed drops to the "beyond" figure.

→ All models on the Apple M3 Max 36GB → LLaVA-CoT specs → Other hardware

FAQ

Can the Apple M3 Max 36GB run LLaVA-CoT?▶

Yes. At Q4 it generates ~33.1-39.9 tok/s and fits up to 151k context; beyond that it spills to SSD swap and slows to ~1.1-1.3 tok/s.

How much context fits?▶

At Q4, up to about 151k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.